Cadence’s CEO discusses consolidation, Moore’s Law, and the next big things.
Lip-Bu Tan, president and CEO of Cadence, sat down with Semiconductor Engineering to talk about consolidation, Moore’s Law, and where the opportunities are in the IoT and automotive markets. What follows are excerpts of that conversation.
SE: What are the big concerns for the semiconductor industry in general, and EDA in particular?
Tan: Top on my list is all the consolidation that’s going on. You have to make sure you’re in the right place to win. Second, chip design complexity is increasing substantially at advanced nodes, which raises concerns about time to market and the need for a whole-system approach. We also need to have more startups because they provide a lot of the breakthroughs and innovation in design. And we need more engineers in this industry. We have a program to recruit every year from colleges and universities. We need new blood, new ideas, and we need to think differently to address new opportunities.
SE: We’ve seen a lot of consolidation lately. Will it continue?
Tan: The trend is unavoidable and is accelerating. Everybody is worried about growth because of the challenges they need to overcome to grow. When you get to 10nm and 7nm and 5nm, those are very expensive nodes. So everyone is looking at how to extend their platform. They’re worried about how to extend the area that they’re in—but not so far that they get into trouble. There also is a concern that the target company has the right culture so it can be integrated in the right location, and that the talent that can be brought on board. You try to obtain efficiency in G&A and market reach and focus. Those are issues you need to address in a maturing market. At the same time, there is a lot of real opportunity in the IoT and automotive. So if you’re looking at these markets, you have to question how you can expand your team.
SE: What are the big challenges for EDA?
Tan: Our customers have very big challenges and needs and we have to provide solutions to them. That provides opportunity for us to grow, but not just with new tools. We have to provide massively parallel solutions for existing tools. Parallelism is required because time to market is critical. On the customer side, if you talk to everyone from the mobile to the cloud guys, power is very important to them along with time to market, and for the IoT guys, cost is going to be critical. They have to bring the cost down and find killer apps. To me this is an exciting time to take on some of these engineering challenges. All of those present a great opportunity for us.
SE: Where are the investments going these days and where will we see new startups?
Tan: There is a lot of opportunity in the data center and cloud area, and in the IoT. I’m a big fan of video, so I’ve been backing a lot of companies doing video. There also is a lot of interest in drones. And another big market is automotive. ADAS (advanced driver assistance systems) is tremendous. These all tie into understanding how the industry is forming. Then you can take whatever solutions are needed, whether it’s IP or tools or PCB or system analysis, and figure out how to connect them all from the system level. That’s a different way of designing.
SE: Is this EDA, or should we start calling it something else?
Tan: EDA is still the core of the business. There is still a lot of room for growth there because it will be required for a lot of challenges that need to be solved. If you look at medical, that’s a tremendous opportunity for growth. Anything that can help prevent heart attacks or identify cancer early enough will rely on a semiconductor process and technology. That can be applied across health care. And it can be applied in other markets, such as automotive. Some day cars will be self-driving so you can get work done and be more productive on the way to the office.
SE: So where exactly does EDA fit into that?
Tan: EDA provides system design enablement. If you look at application development from the standpoint of the systems developer, that drives silicon design. But it’s very expensive to go down to 10nm and 7nm, so you really need first-time success. If you do a re-spin, you miss a market window and it’s very costly. If you look at Ambarella, they’ve done first-time pass. That’s smart engineering. We want to provide the IP and the tools to allow them to do an advanced design that passes the first time. That can be very cost effective. That’s extending from EDA to system design enablement and more application-driven design.
SE: What happens as we move from companies designing for a socket to systems companies that are designing chips for their own systems?
Tan: With a system you have new challenges such as a different power envelope, or you may start with the system level and then look at the chip design. It moves from design to co-design of hardware and software and co-verification. That requires high-level abstractions, and even some of the IP blocks we develop have to be verified at the system level. You’ve got to have the system, the foundry and the constraints in mind, so it’s a complete system-aware design.
SE: What are you hearing back from your customers in terms of Moore’s Law.
Tan: We’ve spent a lot of time meeting with our customers. It’s important to know what they’re designing, but it’s also important to understand what their customers’ customers want, so we understand what the power requirements are, for example. The other part that’s important at advanced nodes is what foundry they should use and what process node they should use. Is power more important, is process more important, or is performance more important? Those are the things we spend a lot of time discussing with partners such as ARM or our foundry partners such as TSMC, GlobalFoundries, Samsung and Intel. We need to understand what they’re doing now, what they need to do in terms of IP enablement, and how to develop a complete solution. Then we can recommend to a customer, ‘For this application and this solution, this foundry will be much better for you.’ We are helping to give them the feedback they need to view us not as a vendor but as a partner. In the end they’ll still make their own decision, but we at least we can give them an honest view that, at this juncture in time and for this design, this is a better tool and a better IP.
SE: How many companies will continue to the most advanced nodes?
Tan: Some of the mixed signal and analog will only go so far. That’s still good business for us. But for some companies that are in graphics and processors and pushing the envelope at the advanced nodes, they will continue to 16, 14 and 10. We are working with them to make sure they can move forward. Others can stay at 65, 40 or 28, and for them FD-SOI may make sense. They have particular features and performance and power that they need. Once they figure out what they need, we can help them tailor a solution and our tools will support them.
SE: Is there still enough volume at the most advanced nodes to warrant the investment in tools?
Tan: Yes. Some of the leading processor and video companies will drive those process nodes. It can be done for a lot less than $250 million, though. I call it smart engineering where you get to first-time pass, not three or four or five revisions.
SE: So you’re bullish enough companies will continue down this path?
Tan: Absolutely. Moore’s Law will continue and will be extended further. We will need to have more innovation, too. I still have to bring my cell phone charger with me after a full day.
SE: One solution that’s been suggested is more parallelism, which has been around for decades. But engineers don’t go into that willingly. Is there a simpler solution?
Tan: You need that parallelism as a tradeoff for the slowing down of performance. When you look at Facebook and Google and Amazon, they realize they need the foundation in silicon that can support killer apps. If you don’t support and invest in that foundation, how are you going to do that, particularly with social media and all the graphics and video. You need to do that using parallelism in the most cost- and power-effective means.
SE: How real is the IoT market for Cadence?
Tan: Everybody is talking about it. A lot of time it’s just connecting to the old devices that you have. But if you look closer, the silicon percentage revenue of those billions or trillions of dollars that have been predicted for the IoT is very tiny. A lot of the value is in the software and services. You really have to look at the services apps and the user experience. Otherwise you’ll throw the device away. To me the killer apps are social media and anything medical. If you can keep track of your blood sugar or your heart, that will be valuable. Those will see very good usage. What’s important to us is the IP business, which is why we bought Tensilica. That IP is programmable. The other part is how to do the design so you can get to market more quickly with low power and low cost. At the end of the day, there are too many startups in the IoT. From my venture capital background, we have to place our bets on who we think will be successful in this.
SE: One other thing that has changed is this really has to be designed for a specific use, rather than a general-purpose processor with a standard configuration, right?
Tan: Very much so. It’s how they’re integrating everything on an SoC and what are the killer applications that they’re driving. How do you bring that performance so the customer is happy enough that they won’t throw it away? The rest can be compromised because it has to fit into a power envelope or a specific size.
SE: You mentioned a couple killer apps, which are glucose and heart monitoring. How close are we to seeing those in the market?
Tan: Some of them are already in the market. But there are new approaches, such as with glucose monitoring, where you don’t have to draw blood to measure sugar. Then there are a couple others such as an EKG. And even more exciting for me is the area of drug discovery. All of those are going to be huge. They involve semiconductors and microfluidics and a lot of other components. This is a great opportunity for semiconductors, and I tell that to students in universities. This is not a sunset industry.
SE: One of the big opportunities these days is in big data. Where do you see Cadence playing?
Tan: Memory and process become very important, and how they all tie in on a system is very important. Another piece of this, and one that I spend a lot of time looking at, is that some of the business models need to be changed. I have three investments I made where they provide data analytics. That’s more valuable, and they can charge a lot more. In the retail space, if you can monitor what’s being sold in real time, you can replace it and sell a lot more, and customers are willing to pay a lot more for the hardware. This requires a big change, and it’s a new one. Then, if the data has more value to the customer, how do you manage that? How do you improve the power and performance? And in the past, you may have only looked at one or two streams of data, when there is much more available. How do you optimize that?
SE: So you’re re-architecting the electronics and the businesses around them?
Tan: That’s correct. And in some of those cases, you can put that data in the cloud so you don’t have to use so much local storage. We are just at the beginning of that—software as a service and the whole cloud infrastructure. It’s a huge opportunity.